Read/write complex types to redshift from spark

I am having difficulty working reading and writing complex types from spark to redshift. I am using the EMR-6.12 image. One point of confusion is that the docs here state that parquet should be used as the temp file format. That does not seem to be supported, see these docs. I also get an error saying parquet is not supported.

Here is a minimal code example:

schema = StructType([
    StructField("name", StringType(), True),
    StructField("hobbies", ArrayType(StringType(), True))])

# Sample data
data = [
    ("Alice", ["reading", "swimming"]),
    ("Bob", ["hiking", "cooking"]),
    ("Charlie", ["gaming", "coding"])
]

df = spark.createDataFrame(data, schema=schema)

# Write 
df.write.format("io.github.spark_redshift_community.spark.redshift").option("url", url).option(
        "dbtable", table_name
    ).option(
        "tempdir", temp_s3
    ).option(  
        "tempformat", "AVRO"  # Throws an error if I use PARQUET
    ).option(
        "forward_spark_s3_credentials", True
    ).mode(
        "error"
    ).mode(
        "overwrite" 
    ).save()

# read it back
df_new = (
        sql_context.read.format("io.github.spark_redshift_community.spark.redshift")
        .option("url", url)
        .option("dbtable",  table_name)
        .option("tempdir", temp_s3)
        .option("forward_spark_s3_credentials", True)
        .schema(schema)
        .load()
)

When I try to do anything with df_new, I get an error about parsing the hobbies column. If I don't specify the schema, the column is read as a string.